Knowledge discovery in databases with categorical data

In this Information Age, rapid, explosive growth of data is an inevitable trend. The flourishing of the Internet makes the situation even worse. Data from many sources are generated simultaneously at a rate that is too fast and too much for people to analyze or make use beyond their transactional purposes. However, people have to cope with this situation under more intense business competition or more urgent scientific needs of discovery.{09}Despite the well-developed database and data warehouse technologies, we strongly need a new way to discover hidden information or “nuggets” inside these databases by taking advantages of today's computing power. This is the goal of a multi-disciplinary field called knowledge discovery in database (KDD).; Besides, the issue of the large amount of data that the knowledge discovery has to sustain, KDD also has to solve the challenges caused by newer or more complicated data types as well. In this dissertation, our main goal is to develop new data transformation techniques that will improve the performance of the current knowledge discovery techniques when applied to categorical databases. Since there are a large number of categorical databases available that share the same special characteristics, the proposed methods will correspond to the special attention that these databases deserve from the knowledge discovery community.; In this dissertation the proposed data transformation methods are applied to several knowledge discovery techniques such as decision trees, Bayesian networks, the nearest neighbor analysis, and the neural networks. The proposed methods are benchmarked with a traffic safety categorical database named the CARE system. The performances of various existing methods and the proposed methodology are also evaluated. It is noted that the proposed methods yield very promising results for the CARE case study.